Place recognition for unmanned vehicle based on 3D LiDAR and semantic point cloud

Place recognition is an important task in the computer vision and robotics communities, with a wild application in many fields. For unmanned vehicle, 3D LiDAR and semantic point cloud is always used for place recognition. Recent state-of-the-art works mostly focus on structural design of the network...

Full description

Saved in:
Bibliographic Details
Main Author: Zhao, Yangyang
Other Authors: Wang Dan Wei
Format: Thesis-Master by Coursework
Language:English
Published: Nanyang Technological University 2024
Subjects:
Online Access:https://hdl.handle.net/10356/173189
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
Description
Summary:Place recognition is an important task in the computer vision and robotics communities, with a wild application in many fields. For unmanned vehicle, 3D LiDAR and semantic point cloud is always used for place recognition. Recent state-of-the-art works mostly focus on structural design of the network. This dissertation introduces relation-based and response-based self-knowledge distillation into the training process and further proposes an instance-to-region supervised knowledge distillation method based on MinkLoc3Dv2 backbone. Experimental evaluation shows excellent performance and generalization on standard benchmarks.